def se_resnet101(num_classes=1_000):
model = ResNet(SEBottleneck, [3, 4, 23, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
out = self.relu(out)
out = self.conv3(out)
out = self.bn3(out)
out = self.se(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=10)
print(model)
def se_resnet18(num_classes=1_000):
model = ResNet(SEBasicBlock, [2, 2, 2, 2], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def se_resnet101(num_classes=1_000):
model = ResNet(SEBottleneck, [3, 4, 23, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def se_resnet18(num_classes=1_000):
model = ResNet(SEBasicBlock, [2, 2, 2, 2], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def resnet34(pretrained=False, model_root=None, **kwargs):
model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
if pretrained:
misc.load_state_dict(model, model_urls['resnet34'], model_root)
return model
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBasicBlock, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def se_resnet50(num_classes=1_000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
model.load_state_dict(
load_state_dict_from_url(
"https://github.com/moskomule/senet.pytorch/releases/download/archive/seresnet50-60a8950a85b2b.pkl"
))
return model
def se_resnet101(num_classes=1_000):
"""Constructs a ResNet-101 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
def prepare_resnet(resnet: ResNet, num_classes: int):
resnet.fc = nn.Linear(resnet.fc.in_features, num_classes)
def train(model: ResNet,
device: torch.device,
train_loader: DataLoader,
val_loader: DataLoader,
arch: str,
epochs: int = 160,
lr: float = 0.1,
momentum: float = 0.9,
bn_only: bool = False) -> dict:
"""
:param bn_only: BN only
:param model: ResNet model
:param device: device to inference
:param train_loader: train loader
:param val_loader: validation loader
:param arch: ResNet<Deep>
:param epochs: number of epochs
:param lr: base learning rate
:param momentum: momentum
:return: dict with statistics
"""
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=lr, momentum=momentum)
lr_scheduler = optim.lr_scheduler.MultiStepLR(optimizer,
milestones=[80, 120])
save_steps = (25, 50, 79, 119)
stats = {"train_loss": [], "val_loss": [], "val_accuracy": []}
def train_():
model.train(True)
total_loss = 0
for i, (inputs, labels) in enumerate(train_loader):
inputs = inputs.to(device)
labels = labels.to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = criterion(outputs, labels)
total_loss += loss.item()
loss.backward()
optimizer.step()
if i % 50 == 0:
print(f"[{epoch + 1}, {i + 1}] loss: {loss.item()}")
epoch_loss = total_loss / len(train_loader)
stats["train_loss"].append(epoch_loss)
def save_():
local_dir = os.path.join("snapshots", arch)
os.makedirs(local_dir, exist_ok=True)
if bn_only:
save_name = os.path.join(local_dir,
f"snapshot_{epoch + 1}_BN_.pth.tar")
else:
save_name = os.path.join(local_dir,
f"snapshot_{epoch + 1}.pth.tar")
torch.save(model.state_dict(), save_name)
print(f"Model {save_name} saved")
def eval_():
model.train(False)
total_loss = 0
total = 0
correct = 0
with torch.no_grad():
for i, (inputs, labels) in enumerate(val_loader):
inputs = inputs.to(device)
labels = labels.to(device)
outputs = model(inputs)
total_loss += criterion(outputs, labels).item()
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
epoch_loss = total_loss / len(val_loader)
epoch_accuracy = correct / total * 100
stats["val_loss"].append(epoch_loss)
stats["val_accuracy"].append(epoch_accuracy)
print(
f"Epoch {epoch + 1}: Accuracy: {epoch_accuracy}. Loss: {epoch_loss}"
)
for epoch in tqdm(range(epochs)):
train_()
print(f"LR: {lr_scheduler.get_last_lr()}")
lr_scheduler.step()
eval_()
if epoch in save_steps:
save_()
print("---------- TRAINING IS FINISHED ----------")
return stats
def Net(num_classes):
model = ResNet(Bottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = torch.nn.AdaptiveAvgPool2d(1)
return model
def se_resnet34(num_classes=1_000):
"""Constructs a ResNet-34 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBasicBlock, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def se_resnet50(num_classes=1_000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
model.load_state_dict(
model_zoo.load_url("https://www.dropbox.com/s/xpq8ne7rwa4kg4c/seresnet50-60a8950a85b2b.pkl"))
return model
def se_resnet101(num_classes=1_000):
"""Constructs a ResNet-101 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 23, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def se_resnet34(num_classes=1000):
model = ResNet(SEBasicBlock, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv3(out)
out = self.bn3(out)
out = self.se(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
def se_resnet50(num_classes=1_000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
model.load_state_dict(model_zoo.load_url("http://data.lip6.fr/cadene/pretrainedmodels/se_resnet50-ce0d4300.pth"))
return model
def __init__(self, NUM_CLASSES):
super(ResNet152, self).__init__()
self.model_name = "ResNet152"
self.model = ResNet(Bottleneck, [3, 8, 36, 3], num_classes=NUM_CLASSES)
#!/usr/bin/env python # encoding: utf-8 """ @Author: yangwenhao @Contact: [email protected] @Software: PyCharm @File: mem_load_test.py @Time: 2020/3/29 10:30 PM @Overview: """ import torch import numpy as np from torchvision.models import ResNet from torchvision.models.resnet import BasicBlock model = ResNet(BasicBlock, [1, 1, 1, 1], num_classes=1000)
def resnet18(**kwargs):
"""Constructs a ResNet-18 model.
"""
model = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def resnet152(**kwargs):
"""Constructs a ResNet-152 model.
"""
model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def resnet34(**kwargs):
"""Constructs a ResNet-34 model.
"""
model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
